Free CCAR-P Solution Design & Architecture Practice Questions

The Solution Design & Architecture domain makes up 17% of the Anthropic CCAR-P exam. Practice free questions covering end-to-end solution design, architecture tradeoffs, component selection, scalability, and system decomposition — each with the correct answer and a detailed explanation. The real exam mixes multiple-choice and multiple-response items; these practice questions use a single-best-answer format.

CCAR-P Solution Design & Architecture Sample Questions with Answers

Sample Question 1 — Solution Design & Architecture

A logistics company demonstrates a Claude prototype that highlights unusual clauses in supplier contracts. Executives want to select a model and automate contract approval immediately. However, legal has not defined acceptable omission rates, which clauses require attorney review, the financial impact of errors, or the current review-time baseline. What should the solution architect recommend FIRST?

  1. A. Select the model with the highest clause-extraction score and estimate operating cost from prototype traffic.
  2. B. Define measurable outcomes, establish the current baseline, and classify decisions by error impact and required legal oversight. (Correct answer)
  3. C. Integrate the prototype with the contract repository and capture attorney corrections during an initial production rollout.
  4. D. Build a multi-stage review workflow that uses separate Claude calls for extraction, risk scoring, and final approval.

Correct answer: B

Explanation: Correct answer (B): The organization must first define success, quantify its current performance, and determine where legal accountability prevents full automation. The explicit absence of omission tolerances, error costs, review boundaries, and a baseline makes architecture or model selection premature. The governing principle is to translate the business problem into measurable requirements before choosing technology. In production, this prevents a compelling demonstration from becoming an uneconomic or unsafe system whose quality cannot be judged. Why the other options are wrong: - Option A: A model comparison will eventually be useful, but an extraction score cannot establish fitness without a representative acceptance threshold or an understanding of error impact. - Option C: Production feedback could improve the system, but deploying before defining approval boundaries and acceptable risk exposes real contracts to an unevaluated process. - Option D: A staged workflow is technically credible, but it commits to an architecture before determining which stages should be automated or retained by legal reviewers.

Sample Question 2 — Solution Design & Architecture

A multinational firm processes 80,000 expense claims each month. Receipts and employee explanations arrive in inconsistent formats, but reimbursement calculations, spending limits, and approval thresholds must be exact and independently auditable. Finance wants Claude to improve processing without increasing manual review of ordinary claims. Which architecture BEST meets these requirements?

  1. A. Use Claude to interpret receipt details and explanations, then apply a deterministic policy engine and route unresolved exceptions to reviewers. (Correct answer)
  2. B. Have Claude interpret each claim and approve it when its generated confidence exceeds a threshold selected by finance.
  3. C. Encode reimbursement policies in the system prompt and ask Claude to return both the calculation and its approval rationale.
  4. D. Use deterministic OCR templates for all receipts and send only claims exceeding approval thresholds to Claude for final decisions.

Correct answer: A

Explanation: Correct answer (A): Claude is well suited to interpreting irregular documents and ambiguous explanations, while exact calculations and approval thresholds belong in deterministic, testable logic. The decisive constraint is that policy enforcement must be exact and independently auditable without forcing ordinary claims into manual review. This separation of probabilistic interpretation from authoritative enforcement limits model uncertainty to the layer that benefits from it. In production, it improves auditability, enables reliable regression testing, and confines human review to genuinely unresolved cases. Why the other options are wrong: - Option B: Confidence-based automation may reduce review volume, but generated confidence is not a deterministic enforcement mechanism and may not be calibrated for financial approval. - Option C: Prompt-encoded policies are convenient to modify, but prompt instructions cannot guarantee exact arithmetic or consistent enforcement of authoritative thresholds. - Option D: Templates can work for stable receipt formats, but the scenario explicitly states that inputs are inconsistent, and Claude would still be placed at the wrong layer as final policy decision-maker.

Sample Question 3 — Solution Design & Architecture

A bank has piloted a service that sends customer complaints to Claude and writes the returned category directly into its case-management platform. It handles 25,000 complaints daily. A wrong category can cause the bank to miss a regulatory response deadline, and operations must trace every escalation decision. Which change is MOST important before production launch?

  1. A. Add dashboards for category volumes, token consumption, model latency, and case-management write failures.
  2. B. Run a second Claude call to critique each category before writing the result into the case platform.
  3. C. Validate structured outputs, apply deadline-aware routing rules, and send invalid or high-risk cases to a traceable review queue. (Correct answer)
  4. D. Fine-tune request prompts using the pilot complaints that received the most detailed customer descriptions.

Correct answer: C

Explanation: Correct answer (C): The pipeline needs an enforcement boundary between probabilistic classification and a regulated downstream system. Because misclassification can cause missed deadlines and every escalation must be traceable, outputs must be structurally validated, checked against deterministic routing requirements, and diverted when invalid or high risk. The architectural principle is to build explicit validation and exception paths before model output changes operational state. In production, this prevents unsafe writes rather than merely observing them after regulatory harm may have occurred. Why the other options are wrong: - Option A: Operational dashboards are valuable, but they are a compensating observability control and do not stop a malformed or unsafe category from entering the case system. - Option B: A critique pass may improve average classification quality, but it remains probabilistic and does not provide deterministic deadline routing or a traceable exception disposition. - Option D: Prompt improvement may reduce some errors, but detailed pilot complaints may not represent production traffic and do not create the required validation and escalation controls.

Sample Question 4 — Solution Design & Architecture

An accounts-payable service extracts fields from 300,000 invoices overnight using Claude. The extraction benchmark is strong, but tests occasionally produce malformed supplier IDs or totals that do not equal line items plus tax. The ledger supports idempotent writes, and all valid invoices must be posted before 6:00 a.m. Which design should the architect select?

  1. A. Post all extracted invoices immediately, then run a reconciliation report that reverses inconsistent ledger entries.
  2. B. Send every extraction to an accounts-payable clerk and post only after the clerk confirms all fields.
  3. C. Increase the extraction prompt's emphasis on supplier-ID formatting and arithmetic before enabling direct posting.
  4. D. Queue extracted records, validate schemas and arithmetic deterministically, and route failures to an exception queue before posting. (Correct answer)

Correct answer: D

Explanation: Correct answer (D): The service should use asynchronous processing to sustain overnight throughput and deterministic checks to prevent malformed identifiers or inconsistent totals from entering the ledger. The explicit constraints are the high batch volume, fixed completion window, and unacceptable downstream data corruption. The governing principle is that structured model output still requires validation at the system boundary. In production, queues absorb processing variability while exception routing preserves automation for valid invoices without compromising ledger integrity. Why the other options are wrong: - Option A: Reconciliation is useful for recovery, but it detects problems after financial records have already been corrupted and creates avoidable reversal work. - Option B: Universal human review would reduce posting risk, but it is unlikely to meet the overnight throughput requirement and is unnecessary for records that pass deterministic checks. - Option C: Prompt improvements may lower the error rate, but they cannot guarantee schema conformance or arithmetic consistency before ledger posting.

Sample Question 5 — Solution Design & Architecture

A public company uses Claude to help prepare quarterly financial commentary. The process always follows four steps: extract approved figures, validate them against the ledger, draft commentary, and verify every claim against the validated figures. Each step consumes an approved artifact from the preceding step, and auditors require reproducible runs. Which architectural pattern is BEST?

  1. A. A sequential workflow with typed intermediate artifacts and a validation gate between each dependent stage. (Correct answer)
  2. B. An autonomous agent that selects tools and decides the next processing step from its intermediate findings.
  3. C. A shared-context multi-agent system in which specialist agents debate until they agree on the final commentary.
  4. D. A single augmented prompt containing the ledger extracts, drafting instructions, validation criteria, and output format.

Correct answer: A

Explanation: Correct answer (A): A sequential workflow directly represents the fixed order and explicit dependencies while preserving each approved artifact for audit and replay. The decisive constraints are reproducibility and the requirement that every stage consume a validated predecessor. The architectural principle is to use deterministic orchestration when control flow is known rather than introducing model-directed autonomy. In production, explicit stage contracts simplify testing, localize failures, and give auditors evidence of how each statement was produced. Why the other options are wrong: - Option B: An autonomous agent is appropriate when the next action must be discovered dynamically, but here the steps and their order are fixed and auditable. - Option C: Specialist debate could improve some drafts, but it introduces coordination overhead and nondeterminism without addressing a need for independent parallel work. - Option D: A single call is operationally simple, but it obscures intermediate validation and weakens reproducibility when every stage requires an approved artifact.

Sample Question 6 — Solution Design & Architecture

An internal assistant receives 120,000 requests daily. About half ask for exact benefit-plan identifiers, 40% ask conceptual questions over policy documents, and 10% request actions such as opening a leave case. Responses should begin within two seconds, and state-changing actions require authorization. Which architecture BEST balances capability, latency, and control?

  1. A. Send every request to one Claude configuration with all policy documents, identifiers, and action tools available in context.
  2. B. Use semantic retrieval for every request, then allow Claude to answer questions or invoke case-management tools as needed.
  3. C. Use a router with deterministic lookup, document-retrieval, and authorized action-workflow paths, plus fallback for uncertain routing. (Correct answer)
  4. D. Deploy separate autonomous agents for benefits, policies, and case management, with a supervisor agent choosing among their responses.

Correct answer: C

Explanation: Correct answer (C): The workload contains three materially different request classes, so routing each class to the appropriate pattern avoids unnecessary context and capability. Exact identifiers should use deterministic lookup, conceptual questions benefit from document retrieval, and state changes require an authorized workflow. The two-second response requirement further disfavors routing everything through a large universal or multi-agent path. In production, workload segmentation reduces latency and cost while preserving a strong control boundary around business actions. Why the other options are wrong: - Option A: A universal configuration is simple and may answer all categories, but carrying irrelevant documents and tools increases context cost, latency, and exposure of action capabilities. - Option B: Semantic retrieval is suitable for conceptual policy questions, but it is inferior for exact identifiers and does not itself enforce authorization for state-changing operations. - Option D: Specialized agents could separate domains, but supervisor coordination adds latency and complexity where deterministic routing and workflows are sufficient.

Sample Question 7 — Solution Design & Architecture

A manufacturer is modernizing a 4-million-line application with tightly coupled modules and limited automated tests. The repository cannot fit usefully into a single model context, production behavior must be preserved, and each deployment must be reversible. A pilot that asked Claude to migrate whole services produced inconsistent interface assumptions. Which approach should the architect recommend?

  1. A. Index the entire repository for semantic retrieval and allow an agent to migrate whichever files appear most relevant to each request.
  2. B. Divide the repository by directory, migrate directories in parallel, and resolve interface inconsistencies during final integration testing.
  3. C. Generate compact summaries of every source file, place all summaries in context, and request a complete modernization patch in one run.
  4. D. Map dependencies first, define bounded migration units and interface contracts, then implement reversible increments with verification gates. (Correct answer)

Correct answer: D

Explanation: Correct answer (D): Dependency mapping must precede implementation because the modules are tightly coupled and the pilot already exposed inconsistent interface assumptions. Bounded units, explicit contracts, verification gates, and reversible increments manage both limited context and production risk. The architectural principle is to decompose complex work into understanding, planning, execution, and verification using durable intermediate artifacts. In production, this limits blast radius, preserves behavioral evidence, and prevents parallel work from silently diverging at shared interfaces. Why the other options are wrong: - Option A: Repository retrieval can provide relevant local context, but relevance alone does not establish migration order, interface contracts, or safe rollback boundaries. - Option B: Parallel migration can accelerate independent work, but directory boundaries do not prove independence, and delaying interface validation repeats the pilot's root failure. - Option C: File summaries reduce context size, but a one-run patch obscures dependencies and creates a large, difficult-to-verify change that conflicts with reversibility.

Sample Question 8 — Solution Design & Architecture

A clinical-review pipeline uses separate Claude calls to extract patient facts, identify guideline criteria, and draft recommendations for clinician approval. Reviewers report that recommendations sometimes rely on facts that cannot be located in the record. The extraction stage currently passes a prose summary downstream, and regulations require traceability to source evidence. Which change MOST directly addresses the problem?

  1. A. Adopt a versioned handoff schema containing source spans, extracted facts, uncertainty markers, and validation status. (Correct answer)
  2. B. Add a final Claude call that checks whether the recommendation sounds consistent with the prose summary.
  3. C. Increase the extraction summary length so downstream stages receive more of the patient's clinical context.
  4. D. Log every prompt and response centrally so reviewers can reconstruct the pipeline after identifying an error.

Correct answer: A

Explanation: Correct answer (A): A structured handoff must preserve provenance and uncertainty at the point where facts are extracted. The explicit requirement is that reviewers trace each recommendation to source evidence, which prose summaries cannot reliably support. The governing principle is to use durable, versioned contracts between probabilistic stages and reject incomplete artifacts rather than asking later stages to infer missing information. In production, this supports validation, replay, schema evolution, and accountable clinician review. Why the other options are wrong: - Option B: A consistency check may detect contradictions with the summary, but it cannot recover source provenance that the extraction stage failed to preserve. - Option C: Longer summaries could retain more context, but they do not create explicit evidence links, uncertainty fields, or machine-verifiable completion states. - Option D: Central logs aid investigation and audit, but they are a compensating control rather than a reliable source-linked contract between pipeline stages.

Keep Practicing

Take the 10-question CCAR-P quick-start test across all 7 domains, or return to the CCAR-P practice test hub for 200+ questions, exam details, and a study plan. Drill the other domains: Integration, Solution Design & Architecture, Evaluation, Testing & Optimization, Governance, Safety & Risk Management, Stakeholder Communication & Lifecycle Management, Claude Models, Prompting & Context Engineering, and Developer Productivity & Operational Enablement.